A multi-source heterogeneous spatio-temporal data management method and system and a readable storage medium

Abstract

The present application relates to a kind of multi-source heterogeneous spatio-temporal data governance method, system and readable storage medium, comprising: the multi-source heterogeneous spatio-temporal data obtained as input data, and the feature analysis of multi-source heterogeneous spatio-temporal data obtains spatio-temporal feature information;Determine the exclusive knowledge base adapted to multi-source heterogeneous spatio-temporal data, each governance rule in exclusive knowledge base is bound with corresponding spatio-temporal metadata label;The spatio-temporal feature information of input data is input to spatio-temporal attention enhanced RAG algorithm model with the spatio-temporal metadata label bound with each governance rule in exclusive knowledge base, based on the dynamic weight of data spatio-temporal characteristic difference self-adaptive adjustment Fusion calculation, obtain comprehensive retrieval score;Based on the obtained each comprehensive retrieval score, recall Top-K target governance rule from exclusive knowledge base;Based on Top-K target governance rule, the uniformity of input data based on spatio-temporal benchmark and the spatio-temporal correlation fusion of multi-source heterogeneous data, obtain the fusion data set after governance.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, and readable storage medium for governing multi-source heterogeneous spatiotemporal data. Background Technology

[0002] Spatiotemporal data governance refers to the process of effectively integrating, cleaning, transforming, storing, analyzing, and applying multi-source heterogeneous spatiotemporal data with time and spatial dimensions through a series of technical means and management strategies. This aims to solve inconsistencies in semantics, format, and spatiotemporal benchmarks, uncover the potential value of the data, and provide high-quality, standardized, and interoperable data support for various spatiotemporal related applications.

[0003] Traditional spatiotemporal data governance methods mainly rely on a single data source or manually configured rule engines. However, it is difficult to cope with the complex relationships of multi-source heterogeneous data such as satellite remote sensing image data, drone aerial photography data, IoT data collected by IoT sensors, geographic information system vector data, and meteorological station observation data through fixed ETL processes. Moreover, the data from different sources have fundamental differences in coordinate systems, time precision, and data formats, which leads to semantic conflicts and spatiotemporal misalignments after data fusion.

[0004] In recent years, although knowledge graphs and RAG technologies have improved data association and retrieval capabilities, static knowledge graphs lack dynamic characterization of spatiotemporal dimensions and cannot effectively express temporal evolution relationships. Furthermore, existing RAG technologies do not fully consider the spatiotemporal data coupling characteristics and lack targeted spatiotemporal attention mechanisms, easily leading to misalignment between retrieved knowledge and spatiotemporal scenarios. Simultaneously, this technology lacks an effective feature space alignment mechanism; the fixed-weight fusion strategy it employs cannot dynamically suppress low-quality data noise, and it fails to form a closed-loop collaborative governance process. This results in insufficient retrieval efficiency and dynamic optimization capabilities, making it difficult to meet the needs of multiple domains for efficient, accurate, and intelligent governance of multi-source heterogeneous spatiotemporal data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a method, system and readable storage medium for multi-source heterogeneous spatiotemporal data governance.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for multi-source heterogeneous spatiotemporal data governance, the method comprising:

[0007] S1. The acquired multi-source heterogeneous spatiotemporal data is used as input data, and the multi-source heterogeneous spatiotemporal data is subjected to feature analysis to obtain spatiotemporal feature information.

[0008] S2. Determine a dedicated knowledge base that adapts to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules, and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags.

[0009] S3. Input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to each governance rule in the exclusive knowledge base into the spatiotemporal attention enhancement RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score.

[0010] S4. Based on the obtained comprehensive search scores, the Top-K target governance rules are retrieved from the exclusive knowledge base by sorting in descending order of rules and filtering by spatiotemporal consistency verification.

[0011] S5. Based on the Top-K target governance rules, the input data is unified in terms of spatiotemporal reference and spatiotemporal correlation fusion of multi-source heterogeneous data is performed to obtain the fused dataset after governance.

[0012] Furthermore, in step S1, the multi-source heterogeneous spatiotemporal data includes satellite remote sensing image data, UAV aerial photography data, Internet of Things (IoT) data collected based on IoT sensors, Geographic Information System (GIS) vector data, and meteorological station observation data.

[0013] Furthermore, in step S2, the dedicated knowledge base adopts a hierarchical storage structure, including a basic rules layer, a spatiotemporal correlation layer, and an adaptation strategy layer, wherein:

[0014] The basic rule layer serves as the underlying support for the exclusive knowledge base, and is used to store and manage rules and information related to spatiotemporal coordinates.

[0015] The spatiotemporal correlation layer, as an intermediate layer, focuses on the intrinsic connections between spatiotemporal data and is used to mine and analyze the temporal correlation rules of spatiotemporal data.

[0016] The adaptation strategy layer serves as the top layer and is used to formulate and implement multi-source data adaptation strategies to achieve the integration and utilization of spatiotemporal data from different sources, in different formats, and with different precision.

[0017] Furthermore, in step S2, when binding the corresponding spatiotemporal metadata tags to each governance rule, the process includes: traversing all governance rules in the exclusive knowledge base. During the traversal, semantic parsing and semantic expansion and reasoning techniques based on domain knowledge graphs are used to match the corresponding spatiotemporal metadata tags for the currently traversed rule, forming an association mapping relationship between "rule content and spatiotemporal metadata".

[0018] Furthermore, in step S3, the dynamic weights adaptively adjusted based on the differences in the spatiotemporal characteristics of the data are fused and calculated to obtain a comprehensive retrieval score, including:

[0019] S31. For the input data and each governance rule, calculate the spatial attention score according to the following formula:

[0020] ;

[0021] in, Indicates spatial attention score, Represents the spatial feature vector of the input data. Represents the learnable input data spatial projection matrix, used to... A linear mapping to a high-dimensional representation space. A spatial metadata vector representing governance rules. Represents a learnable regular space projection matrix, used to... A linear mapping to the same representation space as the input data. Represents the spatial feature dimension. This represents a spatial validity mask, which takes the value of 0 or 1. It is used to determine whether the spatial location of the input data is within the spatial coverage of the governance rules, ensuring that the rule score for spatial mismatch is zero.

[0022] S32. For the input data and each governance rule, calculate the time attention score according to the following formula:

[0023] ;

[0024] in, Indicates time attention score, This represents the time feature vector of the input data, which is obtained by encoding the implicit time information in the input data. Represents the learnable input data time projection matrix, used to... Linear mapping to a high-dimensional representation space, This represents the time-based metadata vector of the governance rules, which is encoded based on the pre-marked time validity period of each rule. This represents the learnable rule-time projection matrix, used to govern the mapping of the original temporal metadata of the rules to the same high-dimensional space as the input data. Representing the time feature dimension, This represents a time validity mask, which takes the value 0 or 1. It is used to determine whether the timestamp of the input data falls within the time validity period of the governance rule, ensuring that the rule with a time mismatch receives a score of zero.

[0025] S33. The spatial attention score and the temporal attention score are weighted and fused according to a dynamic weighting calculation strategy to obtain a comprehensive retrieval score:

[0026] ;

[0027] ;

[0028] in, This represents the overall search score, and k represents the adjustment parameter used to control the sensitivity of weight adjustments. This represents a dynamic weight value, which is dynamically adjusted based on the difference between spatial attention score and temporal attention score.

[0029] Furthermore, in step S4, based on the obtained comprehensive retrieval scores, the Top-K target governance rules are retrieved from the dedicated knowledge base through rule-based descending sorting and spatiotemporal consistency verification, including:

[0030] S41. Based on the obtained comprehensive search scores, sort the rules in the exclusive knowledge base in descending order to generate a candidate rule sequence;

[0031] S42. Traverse each candidate rule in the sequence. During the traversal, if it is determined that the spatial coverage of the current traversal rule does not have a geometric inclusion or intersection relationship with the actual spatial location of the input data, and / or the timestamp of the input data does not fall completely within the time validity period of the current traversal rule, delete the current traversal rule from the sequence.

[0032] S43. Based on the candidate rule sequence left after spatiotemporal verification, select the top K rules from the sequence according to the comprehensive retrieval score from high to low to obtain the Top-K target governance rules.

[0033] Furthermore, in step S5, the process of unifying the spatiotemporal benchmark and fusing the spatiotemporal correlation of multi-source heterogeneous data based on the Top-K target governance rules to obtain the fused dataset after governance includes:

[0034] S51. Traverse the Top-K target governance rules. During the process, perform structured parsing on each target governance rule using a predefined semantic parser.

[0035] S52. Based on the parsing results obtained from the traversal, a rule-parsing field mapping table is constructed in a structured form. The parsing fields include the data field identifier to be identified, the standardized target format, and the associated fusion parameters.

[0036] S53. Based on the rule-parsing field mapping table, the input data is subjected to spatial benchmark unification, time format alignment, and standardized format conversion according to the standardized target format involved therein to obtain a standardized dataset;

[0037] S54. Based on the standardized dataset, according to the association and fusion parameters involved in the rule-parsing field mapping table, multi-source data elements with spatial proximity relationships are associated and bound, and time-series data with different sampling frequencies are aligned to a unified time axis to obtain a spatiotemporally aligned fused dataset.

[0038] Furthermore, the method also includes:

[0039] S6. Based on the spatiotemporally aligned fused dataset, verification is performed from three dimensions: spatial consistency, temporal consistency, and logical consistency. The verified data is then encapsulated, and the encapsulated dataset is output as the governance result.

[0040] Secondly, this application discloses a multi-source heterogeneous spatiotemporal data governance system, the system comprising a spatiotemporal feature analysis module, a dedicated knowledge base adaptation module, a comprehensive retrieval score calculation module, a target governance rule recall module, and a multi-source heterogeneous data association and fusion module, wherein:

[0041] The spatiotemporal feature parsing module is used to take the acquired multi-source heterogeneous spatiotemporal data as input data and perform feature parsing on the multi-source heterogeneous spatiotemporal data to obtain spatiotemporal feature information.

[0042] The dedicated knowledge base adaptation module is used to determine a dedicated knowledge base that adapts to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags.

[0043] The comprehensive retrieval score calculation module is used to input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to various governance rules in the exclusive knowledge base into the spatiotemporal attention enhancement RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score.

[0044] The target governance rule recall module is used to recall the Top-K target governance rules from the dedicated knowledge base based on the obtained comprehensive search scores, through rule descending sorting and spatiotemporal consistency verification.

[0045] The multi-source heterogeneous data association and fusion module is used to unify the spatiotemporal benchmark of the input data and perform spatiotemporal association and fusion of multi-source heterogeneous data based on the Top-K target governance rules to obtain the fused dataset after governance.

[0046] Thirdly, this application discloses a readable storage medium storing a computer program thereon, which, when executed by a processor, implements a multi-source heterogeneous spatiotemporal data governance method as described in any of the preceding claims.

[0047] The beneficial effects of this invention are as follows: First, based on feature parsing, spatiotemporal feature information is extracted, and combined with a dynamic retrieval mechanism (Spatiotemporal Attention Enhancement RAG algorithm model) adapted to a dedicated knowledge base, intelligent matching of governance rules and spatiotemporal characteristics of data can be achieved, effectively solving the problems of rigid rule matching and insufficient retrieval accuracy in traditional methods. Second, by comprehensively ranking the retrieval scores and verifying spatiotemporal consistency, the Top-K rules of recall are ensured to have both authority and scenario adaptability, providing a reliable basis for subsequent processing. Finally, by unifying the spatiotemporal benchmark to eliminate the heterogeneity of multi-source data in coordinate systems and time axes, and by combining spatiotemporal correlation fusion to achieve deep integration of cross-source data in terms of spatial proximity and temporal synchronization, a high-quality fusion dataset with consistent structure and semantic association is finally generated, providing a standardized and reliable data foundation for downstream applications such as spatiotemporal analysis and intelligent decision-making. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a multi-source heterogeneous spatiotemporal data governance method disclosed in this invention;

[0049] Figure 2 A flowchart for matching corresponding spatiotemporal metadata tags using semantic extension and reasoning techniques based on domain knowledge graphs;

[0050] Figure 3 This is a schematic diagram of the structure of a multi-source heterogeneous spatiotemporal data governance system disclosed in this invention;

[0051] Figure 4 This is a schematic diagram of the structure of a readable storage medium disclosed in this invention. Detailed Implementation

[0052] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0053] like Figure 1 As shown, a method for governing multi-source heterogeneous spatiotemporal data includes:

[0054] Step S1: Use the acquired multi-source heterogeneous spatiotemporal data as input data, and perform feature analysis on the multi-source heterogeneous spatiotemporal data to obtain spatiotemporal feature information.

[0055] Specifically, the spatiotemporal feature information includes spatial coordinate features, timestamp features, and data source type features.

[0056] It should be noted that the spatial coordinate features correspond to the geospatial location information of the extracted data records, including latitude and longitude coordinates or planar coordinates. For areal or linear geographic features, this application extracts the coordinates of the smallest bounding rectangle of its spatial extent (i.e., minimum longitude, maximum longitude, minimum latitude, and maximum latitude) as a unified expression of the spatial coordinate features. In one embodiment, if the coordinate system used by the original data is not the target coordinate system, its original coordinate reference system identifier (such as EPSG code) is recorded for use in subsequent spatial benchmark unification. At this stage, only the records are extracted; no coordinate transformation is performed.

[0057] It should be noted that the timestamp feature corresponds to the collection time or time range of the extracted data record. This application will uniformly convert it to the UTC standard time format (such as the ISO8601 standard). For data with time range attributes, this application will also extract its start and end timestamps to form the corresponding time interval feature for use in subsequent steps to determine time validity. In one embodiment, the timestamp precision is automatically adapted according to the data source type.

[0058] It should be noted that the data source type feature is the source identifier of the extracted data record, including the data acquisition platform category (such as satellite remote sensing, drone aerial photography, IoT sensors and GIS databases, etc.), data format category (such as raster data, vector data, point cloud data and tabular data, etc.), and the business scenario category. These three types of identifiers together constitute the data source type feature, which is used for accurate matching of multi-source data adaptation strategies in subsequent steps.

[0059] In one embodiment, after the extraction of spatiotemporal feature information is completed, the above three types of spatiotemporal feature information will be attached to the original data record in a structured form to form data entries carrying spatiotemporal feature tags for use in subsequent steps.

[0060] Step S2: Determine a dedicated knowledge base adapted to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules, and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags.

[0061] Specifically, the spatiotemporal metadata tags include spatial dimension tags for marking spatial coverage, time dimension tags for marking the effective time period, and data type tags for marking the applicable data types.

[0062] It should be noted that the dedicated knowledge base constructed here includes spatiotemporal coordinate standards, temporal association rules, and multi-source data adaptation strategies. Specifically, this application will tag each governance rule in the knowledge base with spatiotemporal metadata to clearly define the spatial coverage, time validity period, and applicable data types of each rule.

[0063] In one embodiment, this application specifies the spatial coverage of each rule in the form of a minimum bounding rectangle or polygon vector, defines its geographical boundaries, and records the corresponding coordinate reference system identifier (CRS identifier, such as EPSG code) to ensure that the expression of spatial coverage has a clear coordinate reference system basis. Furthermore, for globally applicable rules (such as UTC time format conversion rules), their spatial coverage is marked as global.

[0064] In one embodiment, this application defines the validity period of each rule in the form of a UTC timestamp, including the effective start time and the expiration time, to ensure that expired or ineffective governance rules are not recalled during the retrieval phase. Additionally, for universally applicable rules that are valid for a long period, their expiration time is marked as empty (indicating indefinite validity).

[0065] In one embodiment, this application declares the data format category (such as raster data, vector data, time series table data, etc.) and business scenario tags (such as land use, meteorology and hydrology, smart city, etc.) applicable to each governance rule, so as to support rule filtering based on data type during subsequent retrieval.

[0066] Step S3: Input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to each governance rule in the exclusive knowledge base into the spatiotemporal attention enhancement RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score.

[0067] Specifically, the algorithm model first calculates the spatial attention score and temporal attention score of the input data and each governance rule respectively. Then, based on the difference between the two scores, it controls the sensitivity by adjusting parameters. After calculating the dynamic weights, it obtains the comprehensive retrieval score corresponding to each governance rule by weighted summation.

[0068] Step S4: Based on the obtained comprehensive search scores, the Top-K target governance rules are retrieved from the dedicated knowledge base by sorting in descending order of rules and filtering by spatiotemporal consistency verification.

[0069] First, this application performs a preliminary sorting of all rules in its proprietary knowledge base based on comprehensive search scores using a pre-defined sorting algorithm. This algorithm considers scores from highest to lowest, ensuring that rules with higher scores are ranked higher. After the preliminary sorting, a spatiotemporal consistency verification step is performed. For each rule in the preliminary sorting, this application checks whether its applicable spatiotemporal scope matches the spatiotemporal information of the current governance scenario. Finally, after the spatiotemporal consistency verification, the remaining rules are reorganized according to the preliminary sorting order, and the top K rules are selected as target governance rules for recall. These rules will provide a solid and reliable basis for subsequent governance decisions and execution, helping to improve governance efficiency and effectiveness while reducing governance costs and risks.

[0070] Step S5: Based on the Top-K target governance rules, the input data is unified in terms of spatiotemporal reference and spatiotemporal correlation fusion of multi-source heterogeneous data to obtain the fused dataset after governance.

[0071] Specifically, based on the Top-K target governance rules of parsing and recall, this application will automatically identify, standardize and transform (such as coordinate system one, time format alignment) and associate and fuse (such as spatial topology association, multi-source time series alignment) the current input data according to the rule content, so as to realize the alignment and integration of data in the spatiotemporal dimension.

[0072] Finally, this application will encapsulate and output the data after standardization and correlation fusion processing according to a unified spatiotemporal data structure. At the same time, this application will also attach a processing traceability label to each output data and record the governance rule ID it follows, the sequence of operations executed, and the quality confidence score, ultimately forming a high-quality standardized dataset that can be directly used for subsequent spatiotemporal analysis and intelligent decision-making.

[0073] As can be seen from the above, the multi-source heterogeneous spatiotemporal data governance method disclosed in this application firstly extracts spatiotemporal feature information based on feature parsing, and combines it with a dynamic retrieval mechanism (spatiotemporal attention-enhanced RAG algorithm model) adapted to a dedicated knowledge base, which can realize intelligent matching between governance rules and spatiotemporal characteristics of data, effectively solving the problems of rigid rule matching and insufficient retrieval accuracy in traditional methods. Secondly, by comprehensively sorting the retrieval scores and verifying spatiotemporal consistency, it ensures that the recalled Top-K rules have both authority and scenario adaptability, providing a reliable basis for subsequent processing. Finally, by unifying the spatiotemporal benchmark to eliminate the heterogeneity of multi-source data in coordinate system and time axis, and by combining spatiotemporal correlation fusion to achieve deep integration of cross-source data in terms of spatial proximity and temporal synchronization, a high-quality fusion dataset with consistent structure and semantic association is finally generated, providing a standardized and reliable data foundation for downstream applications such as spatiotemporal analysis and intelligent decision-making.

[0074] In one embodiment, in step S1, the multi-source heterogeneous spatiotemporal data includes satellite remote sensing image data, UAV aerial photography data, Internet of Things (IoT) data collected based on IoT sensors, Geographic Information System (GIS) vector data, and meteorological station observation data.

[0075] It should be noted that multi-source heterogeneous spatiotemporal data refers to a collection of data with spatial location information and temporal attributes that originate from different acquisition devices, platforms, and formats. These data differ significantly in spatial reference frame, temporal resolution, data format, and semantic representation, making them typical heterogeneous data. Direct fusion of such data would lead to semantic conflicts and spatiotemporal misalignment.

[0076] In one embodiment, the aforementioned multi-source heterogeneous spatiotemporal data also includes structured and semi-structured data. Structured data mainly includes tabular data bound to spatiotemporal information, such as monitoring data tables with spatial coordinates and timestamps, equipment operation and maintenance ledgers, land ownership information tables, and business statistical reports, with standardized data fields and uniform formats. Semi-structured data mainly includes JSON / XML data with spatiotemporal tags, interface messages, equipment logs, trajectory records, and inspection forms, etc. Although they do not have a fixed table structure, they contain parsable spatiotemporal and business attributes. This application does not limit the specific data types.

[0077] In one embodiment, in step S2, the dedicated knowledge base adopts a hierarchical storage structure, including a basic rules layer, a spatiotemporal correlation layer, and an adaptation strategy layer, wherein:

[0078] The basic rule layer serves as the underlying support for the dedicated knowledge base, and is used to store and manage rules and information related to spatiotemporal coordinates.

[0079] Specifically, the basic rules layer covers the definition information, transformation parameters and transformation methods of various coordinate systems and elevation datums, including transformation parameters (seven parameters, four parameters, etc.) of mainstream coordinate systems such as WGS84 and CGCS2000, coordinate system identification rules (identifying the coordinate system used in the current data based on EPSG code or coordinate system name), and elevation datum definitions (including the 1985 National Elevation Datum, EGM2008 geoid model, etc.).

[0080] The spatiotemporal correlation layer serves as an intermediate layer, focusing on the intrinsic connections between spatiotemporal data and used to mine and analyze the temporal correlation rules of spatiotemporal data.

[0081] Specifically, the spatiotemporal correlation layer includes the correlation and alignment specifications of spatiotemporal data in the time dimension, including: timestamp alignment accuracy rules (defining the timestamp accuracy requirements and alignment methods corresponding to different data source types), sampling frequency standardization rules (for multi-source data with inconsistent sampling frequencies, describing the applicable conditions and operating procedures for time series alignment using spatiotemporal interpolation algorithms such as linear interpolation, nearest neighbor interpolation, and spline interpolation), and time series validity judgment rules (defining the data time validity window according to the business scenario; data exceeding the window is considered expired data and triggers the corresponding processing flow).

[0082] The adaptation strategy layer serves as the top layer and is used to formulate and implement multi-source data adaptation strategies to achieve the integration and utilization of spatiotemporal data from different sources, in different formats, and with different precision.

[0083] Specifically, the adaptation strategy layer covers the formulation of format conversion, quality verification, and semantic mapping rules for data from different sources, such as satellite remote sensing image data, UAV aerial photography data, IoT data collected by IoT sensors, GIS vector data, and meteorological station observation data.

[0084] It should be noted that this application clearly defines the spatial coverage of the rules in the spatial dimension, which specifically defines the geographic boundary with the smallest bounding rectangle or polygon vector and records the coordinate reference system identifier. In the time dimension, this application also marks the time validity period of the rules and defines the start and end times of effectiveness with UTC timestamps. In the data type dimension, this application also declares the applicable data format and business scenario labels for the rules.

[0085] In one embodiment, please refer to Figure 2 In step S2, when binding the corresponding spatiotemporal metadata tags to each governance rule, the process includes: traversing all governance rules in the exclusive knowledge base. During the traversal, semantic parsing and semantic expansion and reasoning techniques based on domain knowledge graphs are used to match the corresponding spatiotemporal metadata tags for the current traversed rule, forming an association mapping relationship between "rule content and spatiotemporal metadata".

[0086] Specifically, during the semantic parsing process, this application utilizes the semantic expansion function of knowledge graphs to enrich and expand the semantics of entities in the rules (such as location, time, and event) to better uncover potential relationships and implicit information between entities. Simultaneously, this application also employs the reasoning mechanism of knowledge graphs to infer the spatiotemporal metadata tags that the current rule may involve, based on predefined spatiotemporal association rules and patterns in the knowledge graph, after uncovering the potential relationships and implicit information between entities, thereby improving the accuracy and comprehensiveness of the matching.

[0087] In one embodiment, step S3, where the dynamic weights adaptively adjusted based on differences in data spatiotemporal characteristics are fused to obtain a comprehensive retrieval score, includes:

[0088] Step S31: For the input data and each governance rule, calculate the spatial attention score according to the following formula:

[0089] ;

[0090] in, Indicates spatial attention score, Represents the spatial feature vector of the input data. Represents the learnable input data spatial projection matrix, used to... A linear mapping to a high-dimensional representation space. A spatial metadata vector representing governance rules. Represents a learnable regular space projection matrix, used to... A linear mapping to the same representation space as the input data. Represents the spatial feature dimension. This represents a spatial validity mask, which takes the value 0 or 1. It is used to determine whether the spatial location of the input data is within the spatial coverage of the governance rules, ensuring that rules with spatial mismatches receive a score of zero.

[0091] Step S32: For the input data and each governance rule, calculate the time attention score according to the following formula:

[0092] ;

[0093] in, Indicates time attention score, This represents the time feature vector of the input data, which is obtained by encoding the implicit time information in the input data. Represents the learnable input data time projection matrix, used to... Linear mapping to a high-dimensional representation space, This represents the time-based metadata vector of the governance rules, which is encoded based on the pre-marked time validity period of each rule. This represents the learnable rule-time projection matrix, used to govern the mapping of the original temporal metadata of the rules to the same high-dimensional space as the input data. Representing the time feature dimension, This represents a time validity mask, which takes the value 0 or 1. It is used to determine whether the timestamp of the input data falls within the time validity period of the governance rule, ensuring that the rule with a time mismatch receives a score of zero.

[0094] Step S33: The spatial attention score and the temporal attention score are weighted and fused according to a dynamic weighting calculation strategy to obtain the comprehensive retrieval score.

[0095] ;

[0096] ;

[0097] in, This represents the overall search score, and k represents the adjustment parameter used to control the sensitivity of weight adjustments. This represents a dynamic weight value, which is dynamically adjusted based on the difference between spatial attention score and temporal attention score.

[0098] Specifically, dynamic weights The calculation is specifically based on minus The product of the difference and the adjustment parameter k is used as input. The adjustment parameter k controls the function's sensitivity to changes in the difference. Based on the characteristics of this function, The weighting will dynamically fluctuate between 0 and 1 based on the difference between spatial and temporal attention scores. When the difference between the two scores is large, the weighting will favor the higher-scoring candidate; when the difference is small, the weighting will be relatively balanced. Finally, through dynamic weighting... By multiplying the spatial attention score and the temporal attention score separately, and then summing the results, a weighted summation is obtained to obtain the comprehensive retrieval score that takes into account both spatial and temporal attention factors. This provides more reasonable and comprehensive data support for relevant decision-making.

[0099] In one embodiment, in step S4, based on the obtained comprehensive retrieval scores, the Top-K target governance rules are retrieved from the dedicated knowledge base through rule-based descending sorting and spatiotemporal consistency verification, including:

[0100] Step S41: Based on the obtained comprehensive search scores, sort the rules in the exclusive knowledge base in descending order to generate a candidate rule sequence.

[0101] Specifically, this application will sort the rules in the exclusive knowledge base in descending order according to the comprehensive search score, thereby constructing a candidate rule sequence that reflects the priority order of the rules when processing the current data.

[0102] In one embodiment, the present application will also pre-set a score filtering threshold, and before sorting in descending order, remove low-relevance rules whose comprehensive search score is lower than the score filtering threshold, so as to ensure that the subsequent descending sorting and candidate rule sequence generation process is only carried out for high-relevance rules, thereby improving the accuracy and efficiency of rule screening.

[0103] Step S42: Traverse each candidate rule in the sequence. During the traversal, if it is determined that the spatial coverage of the current traversal rule does not have a geometric inclusion or intersection relationship with the actual spatial location of the input data, and / or the timestamp of the input data does not fall completely within the time validity period of the current traversal rule, delete the current traversal rule from the sequence.

[0104] Specifically, for each candidate rule in the sequence, it needs to be judged from two key dimensions: space and time. Spatially, a candidate rule is only applicable to the input data if its spatial coverage is geometrically contained within or intersects with the actual spatial location of the input data. Temporally, the timestamp of the input data must fall entirely within the effective time period of the candidate rule to ensure its temporal applicability. Only when the currently traversed rule satisfies all the above conditions—that is, it matches both spatially and temporally—is it considered to have a strong spatiotemporal correlation with the input data and can be effectively applied to the current input data. Therefore, it is retained from the sequence to ensure that the rule set used in subsequent processing has higher accuracy and effectiveness.

[0105] Step S43: Based on the candidate rule sequence left after spatiotemporal verification, select the top K rules from the sequence according to the comprehensive retrieval score from high to low to obtain the Top-K target governance rules.

[0106] Specifically, this application selects rules sequentially from the head of the sequence based on a pre-set K value until K rules are selected, resulting in the Top - K target governance rules. The K value can be flexibly adjusted according to the data complexity and the size of the knowledge base, and this application does not impose any restrictions on it.

[0107] In one embodiment, for the Top-K target governance rules of the recall, this application will attach a corresponding comprehensive search score to them and use the specific value of the comprehensive search score as a priority identifier to ensure that the governance rule with the higher score is used first when resolving multiple rule conflicts or executing the rule chain in the future, thereby ensuring the optimal adaptation of the governance result in the spatiotemporal dimension.

[0108] In one embodiment, step S5 involves unifying the spatiotemporal benchmark and fusing the spatiotemporal correlation of multi-source heterogeneous data based on the Top-K target governance rules to obtain a fused dataset after governance, including:

[0109] Step S51: Traverse the Top-K target governance rules. During this process, each target governance rule is structured and parsed using a predefined semantic parser.

[0110] It should be noted that a semantic parser can be understood as a rule parsing engine based on natural language processing (NLP) or domain-specific language (DSL). Its core functions include: (1) analyzing and decomposing the conditions, actions and constraints in the rules through syntax tree analysis; (2) using entities (e.g., field-level entities, format-level entities and operation-level entities) to identify and extract key elements (such as field names, target formats, and related fields); and (3) combining intent classification to determine the rule type (such as standardization, deduplication, and association fusion).

[0111] Specifically, the semantic parser will parse the data field identifiers to be identified (i.e., the fields to be processed involved in the explicit rules), the standardized target format (i.e. the target data specifications after field conversion, including spatial coordinate system, time representation, etc.) and the association and fusion parameters (if the rule involves multi-field association or fusion operations, parameters such as association conditions, weights or calculation logic will be extracted).

[0112] Step S52: Based on the parsing results obtained from the traversal, a rule-parsing field mapping table is constructed in a structured form. The parsing fields include the data field identifier to be identified, the standardized target format, and the associated fusion parameters.

[0113] Specifically, this application will sort the rule execution plan according to the category of governance operation and the logical dependency of rule execution, that is, the operation of setting coordinate system one and time format alignment will be executed first, data standardization transformation will be executed second, and correlation and fusion operation will be executed last.

[0114] Step S53: Based on the rule-parsing field mapping table, the input data is subjected to spatial benchmark unification, time format alignment, and standardized format conversion according to the standardized target format involved therein to obtain a standardized dataset.

[0115] Specifically, during the process of unifying the spatial reference system, this application identifies the original spatial coordinate information in the input data. If it is determined that the coordinate reference system is inconsistent with the target coordinate system specified in the rules, a coordinate transformation is performed using a grid offset algorithm to generate new coordinate fields under the unified spatial reference system. Specifically, this application calls a preset coordinate transformation function according to the coordinate system transformation rules and corresponding execution parameters to transform the spatial coordinates of the input data from the original coordinate system to the unified target coordinate system. After the transformation is completed, this application further updates the spatial coordinate characteristics and coordinate reference system identifier of the data.

[0116] Specifically, during the time format alignment process, this application parses the timestamp strings in the input data and converts them into a unified UTC time format. Then, according to the rules, it interpolates and fills in missing time values ​​or removes outliers to generate a standardized time field. Specifically, this application uses the time format unification rules in the time series association rule layer to convert the timestamps of the input data from their original format to the standard UTC time format. For data with insufficient time resolution, this application completes the timestamp precision according to the default completion strategy defined in the rules.

[0117] Specifically, during the standardization format conversion process, operations such as format conversion (converting input data from the original format to a unified internal storage format) and resolution resampling (resampling data with inconsistent spatial resolution or sampling frequency according to the rules covered in the adaptation strategy layer) are performed in sequence to complete the standardization conversion and ensure that all input data have a unified spatiotemporal reference.

[0118] Step S54: Based on the standardized dataset, according to the association and fusion parameters involved in the rule-parsing field mapping table, multi-source data elements with spatial proximity relationships are associated and bound, and time-series data with different sampling frequencies are aligned to a unified time axis to obtain a spatiotemporally aligned fused dataset.

[0119] Specifically, in the process of spatial topological association, this application will associate and bind multi-source data points with spatial proximity relationships through spatial inclusion queries or distance threshold calculations. Specifically, for multi-source spatial data that has achieved spatial benchmark unification, this application will calculate the spatial topological relationships (including intersection, inclusion, adjacency, etc.) between data elements according to the spatial topological association strategy in the association fusion parameters, establish a spatial topological association index, and associate and bind multi-source data elements with spatial topological relationships to form a fused dataset with a consistent spatial context.

[0120] Specifically, in the process of multi-source time series alignment, this application employs a dynamic time warping algorithm to align data sequences with different sampling frequencies onto a unified time axis. Specifically, for multi-source time series data that has completed time format alignment, this application uses the spatiotemporal interpolation algorithm specified in the spatiotemporal correlation layer, based on the multi-source time series alignment strategy in the correlation fusion parameters, to interpolate and resample the time series data with different sampling frequencies, unifying all data sequences to the target time step and achieving accurate alignment of the multi-source time series data on the time axis.

[0121] In one embodiment, to address the inconsistency in feature distribution caused by differences in resolution, sampling frequency, and imaging principles among different modalities such as satellite remote sensing, UAV aerial photography, IoT sensors, and GIS databases, this application introduces an adversarial domain alignment mechanism during the association and fusion stage. Through adversarial training between the domain discriminator network and the feature extractor network, domain-invariant feature representations are learned, mitigating the differences in feature distribution among multi-source data. Furthermore, to address the issue of low-quality data in some modalities due to environmental interference such as cloud cover and sensor malfunctions, this application adaptively adjusts the weight of each modality in the fusion process based on its quality confidence level, reducing noise interference from low-quality modalities and improving the accuracy and robustness of multi-source data fusion.

[0122] In one embodiment, the method further includes:

[0123] Step S6: Based on the spatiotemporally aligned fused dataset, verification is performed from three dimensions: spatial consistency, temporal consistency, and logical consistency. The verified data is then encapsulated, and the encapsulated dataset is output as the governance result.

[0124] Specifically, for spatial consistency verification, this application checks whether the coordinate reference system identifiers of all data entries in the spatiotemporally aligned fused dataset are unified to the target coordinate system, which is achieved by reading and comparing each entry one by one. In particular, this application also performs a reverse coordinate transformation on randomly sampled data entries (i.e., transforming the coordinate values ​​in the target coordinate system back to the original coordinate system), calculates the deviation between the reverse transformation result and the original coordinate values, and if the deviation exceeds a preset coordinate accuracy threshold, the corresponding data entry is marked as having a coordinate transformation anomaly and a reprocessing procedure is triggered to ensure the accuracy of spatial benchmark unification.

[0125] Specifically, for time consistency verification, this application verifies whether the timestamps of all data entries in the spatiotemporally aligned fused dataset conform to the UTC standard time format. Specifically, for time-series data, this application further verifies whether the time step of the time series is consistent with the target time step. By calculating the difference between adjacent timestamps and comparing it with the target time step, data points exceeding a preset tolerance threshold are marked as time alignment anomalies and trigger a reprocessing procedure to ensure the accuracy of time format alignment.

[0126] Specifically, for logical consistency verification, this application examines the logical correctness of each data entry in the spatiotemporally aligned fused dataset in terms of attribute field integrity, spatial topological association integrity, and multi-source temporal alignment integrity. Specifically, attribute field integrity verification checks whether all required fields have values, whether the unified field naming system is correct, and whether missing values ​​have been handled according to rules; spatial topological association integrity verification checks for the existence of isolated data elements without any established associations; and multi-source temporal alignment integrity verification checks for the existence of entire missing columns in the fused temporal matrix. For logical conflicts discovered during the verification process, conflict resolution is performed according to the rule priority markers determined in step S601. The results of rule governance with higher priority are retained, while the results of rule governance with lower priority are overwritten.

[0127] After completing the consistency verification of the above three dimensions, this application will also summarize and judge the verification results: If all verification dimensions pass (i.e., there are no abnormal data entries or abnormal entries have been reprocessed), the verified data will be packaged according to a unified spatiotemporal data model, outputting a standardized and integrated spatiotemporal dataset, and attaching complete verification traceability information (including total data volume, statistics of various governance operations, verification pass rate, and governance rule reference records, etc.), forming a closed loop of collaborative governance from data access, rule matching, intelligent governance to consistency verification, providing a high-quality and reliable data foundation for spatiotemporal data analysis and intelligent decision-making in the fields of land and space governance and smart cities; If there are data entries that still fail verification after reprocessing, this part of the data will be output separately as an abnormal dataset with a detailed explanation of the reasons for the abnormality, and will be manually reviewed or processed in subsequent iterations to ensure the overall data quality of the main output dataset is reliable, and complete the intelligent governance of multi-source heterogeneous spatiotemporal data.

[0128] Please refer to Figure 3 This application discloses a multi-source heterogeneous spatiotemporal data governance system, which includes a spatiotemporal feature analysis module, a dedicated knowledge base adaptation module, a comprehensive retrieval score calculation module, a target governance rule recall module, and a multi-source heterogeneous data association and fusion module, wherein:

[0129] The spatiotemporal feature parsing module is used to take the acquired multi-source heterogeneous spatiotemporal data as input data and perform feature parsing on the multi-source heterogeneous spatiotemporal data to obtain spatiotemporal feature information.

[0130] The dedicated knowledge base adaptation module is used to determine a dedicated knowledge base that is adapted to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags.

[0131] The comprehensive retrieval score calculation module is used to input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to various governance rules in the exclusive knowledge base into the spatiotemporal attention-enhanced RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score.

[0132] The target governance rule recall module is used to recall the Top-K target governance rules from the dedicated knowledge base based on the obtained comprehensive search scores, through rule descending sorting and spatiotemporal consistency verification.

[0133] The multi-source heterogeneous data association and fusion module is used to unify the spatiotemporal benchmark of the input data and perform spatiotemporal association and fusion of multi-source heterogeneous data based on the Top-K target governance rules to obtain the fused dataset after governance.

[0134] In one embodiment, the above modules are also used to implement a multi-source heterogeneous spatiotemporal data governance method as described in any of the preceding claims, and this application does not limit this.

[0135] As can be seen from the above, the multi-source heterogeneous spatiotemporal data governance system disclosed in this application firstly extracts spatiotemporal feature information based on feature parsing, and combines it with a dynamic retrieval mechanism (spatiotemporal attention-enhanced RAG algorithm model) adapted to a dedicated knowledge base, which can realize intelligent matching between governance rules and spatiotemporal characteristics of data, effectively solving the problems of rigid rule matching and insufficient retrieval accuracy in traditional methods. Secondly, by comprehensively sorting the retrieval scores and verifying spatiotemporal consistency, it ensures that the recalled Top-K rules are both authoritative and scenario-adaptable, providing a reliable basis for subsequent processing. Finally, by unifying the spatiotemporal benchmark to eliminate the heterogeneity of multi-source data in coordinate system and time axis, and by combining spatiotemporal correlation fusion to achieve deep integration of cross-source data in terms of spatial proximity and temporal synchronization, a high-quality fusion dataset with consistent structure and semantic association is finally generated, providing a standardized and reliable data foundation for downstream applications such as spatiotemporal analysis and intelligent decision-making.

[0136] Please refer to Figure 4 This application discloses a readable storage medium storing a computer program thereon, which, when executed by a processor, implements a multi-source heterogeneous spatiotemporal data governance method as described in any of the preceding claims.

[0137] As can be seen from the above, the readable storage medium disclosed in this application firstly extracts spatiotemporal feature information based on feature parsing, and combines it with a dynamic retrieval mechanism (spatiotemporal attention-enhanced RAG algorithm model) adapted to a dedicated knowledge base, which can realize intelligent matching between governance rules and data spatiotemporal characteristics, effectively solving the problems of rigid rule matching and insufficient retrieval accuracy in traditional methods. Secondly, by comprehensively ranking the retrieval scores and verifying spatiotemporal consistency, it ensures that the recalled Top-K rules are both authoritative and scenario-adaptable, providing a reliable basis for subsequent processing. Finally, by unifying the spatiotemporal benchmark to eliminate the heterogeneity of multi-source data in coordinate systems and time axes, and by combining spatiotemporal correlation fusion to achieve deep integration of cross-source data in terms of spatial proximity and temporal synchronization, a high-quality fusion dataset with consistent structure and semantic association is finally generated, providing a standardized and reliable data foundation for downstream applications such as spatiotemporal analysis and intelligent decision-making.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for governing multi-source heterogeneous spatiotemporal data, characterized in that, The method includes: S1. The acquired multi-source heterogeneous spatiotemporal data is used as input data, and the multi-source heterogeneous spatiotemporal data is subjected to feature analysis to obtain spatiotemporal feature information. S2. Determine a dedicated knowledge base that adapts to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules, and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags. S3. Input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to each governance rule in the exclusive knowledge base into the spatiotemporal attention enhancement RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score. S4. Based on the obtained comprehensive search scores, the Top-K target governance rules are retrieved from the exclusive knowledge base by sorting in descending order of rules and filtering by spatiotemporal consistency verification. S5. Based on the Top-K target governance rules, the input data is unified in terms of spatiotemporal reference and spatiotemporal correlation fusion of multi-source heterogeneous data is performed to obtain the fused dataset after governance; In step S3, the dynamic weights adaptively adjusted based on the differences in spatiotemporal characteristics of the data are fused and calculated to obtain a comprehensive retrieval score, including: S31. For the input data and each governance rule, calculate the spatial attention score according to the following formula: ； in, Indicates spatial attention score, Represents the spatial feature vector of the input data. Represents the learnable input data spatial projection matrix, used to... A linear mapping to a high-dimensional representation space. A spatial metadata vector representing governance rules. Represents a learnable regular space projection matrix, used to... A linear mapping to the same representation space as the input data. Represents the spatial feature dimension. This represents a spatial validity mask, which takes the value of 0 or 1. It is used to determine whether the spatial location of the input data is within the spatial coverage of the governance rules, ensuring that the rule score for spatial mismatch is zero. S32. For the input data and each governance rule, calculate the time attention score according to the following formula: ； in, Indicates time attention score, This represents the time feature vector of the input data, which is obtained by encoding the implicit time information in the input data. Represents the learnable input data time projection matrix, used to... Linear mapping to a high-dimensional representation space, This represents the time-based metadata vector of the governance rules, which is encoded based on the pre-marked time validity period of each rule. This represents the learnable rule-time projection matrix, used to govern the mapping of the original temporal metadata of the rules to the same high-dimensional space as the input data. Representing the time feature dimension, This represents a time validity mask, which takes the value 0 or 1. It is used to determine whether the timestamp of the input data falls within the time validity period of the governance rule, ensuring that the rule with a time mismatch receives a score of zero. S33. The spatial attention score and the temporal attention score are weighted and fused according to a dynamic weighting calculation strategy to obtain a comprehensive retrieval score: ； ； in, This represents the overall search score, and k represents the adjustment parameter used to control the sensitivity of weight adjustments. This represents a dynamic weight value, which is dynamically adjusted based on the difference between spatial attention score and temporal attention score.

2. The method according to claim 1, characterized in that, In step S1, the multi-source heterogeneous spatiotemporal data includes satellite remote sensing image data, UAV aerial photography data, Internet of Things (IoT) data collected based on IoT sensors, Geographic Information System (GIS) vector data, and meteorological station observation data.

3. The method according to claim 1, characterized in that, In step S2, the dedicated knowledge base adopts a hierarchical storage structure, including a basic rules layer, a spatiotemporal correlation layer, and an adaptation strategy layer, wherein: The basic rule layer serves as the underlying support for the exclusive knowledge base, and is used to store and manage rules and information related to spatiotemporal coordinates. The spatiotemporal correlation layer, as an intermediate layer, focuses on the intrinsic connections between spatiotemporal data and is used to mine and analyze the temporal correlation rules of spatiotemporal data. The adaptation strategy layer serves as the top layer and is used to formulate and implement multi-source data adaptation strategies to achieve the integration and utilization of spatiotemporal data from different sources, in different formats, and with different precision.

4. The method according to claim 1, characterized in that, In step S2, when binding the corresponding spatiotemporal metadata tags to each governance rule, the process includes: traversing all governance rules in the exclusive knowledge base. During the traversal, semantic parsing and semantic expansion and reasoning techniques based on domain knowledge graphs are used to match the corresponding spatiotemporal metadata tags for the currently traversed rule, forming an association mapping relationship between "rule content and spatiotemporal metadata".

5. The method according to claim 1, characterized in that, In step S4, based on the obtained comprehensive retrieval scores, the Top-K target governance rules are retrieved from the dedicated knowledge base through rule-based descending sorting and spatiotemporal consistency verification, including: S41. Based on the obtained comprehensive search scores, sort the rules in the exclusive knowledge base in descending order to generate a candidate rule sequence; S42. Traverse each candidate rule in the sequence. During the traversal, if it is determined that the spatial coverage of the current traversal rule does not have a geometric inclusion or intersection relationship with the actual spatial location of the input data, and / or the timestamp of the input data does not fall completely within the time validity period of the current traversal rule, delete the current traversal rule from the sequence. S43. Based on the candidate rule sequence left after spatiotemporal verification, select the top K rules from the sequence according to the comprehensive retrieval score from high to low to obtain the Top-K target governance rules.

6. The method according to claim 1, characterized in that, In step S5, the input data is subjected to spatiotemporal benchmark unification and spatiotemporal correlation fusion of multi-source heterogeneous data based on the Top-K target governance rules to obtain the fused dataset after governance, including: S51. Traverse the Top-K target governance rules. During the process, perform structured parsing on each target governance rule using a predefined semantic parser. S52. Based on the parsing results obtained from the traversal, a rule-parsing field mapping table is constructed in a structured form. The parsing fields include the data field identifier to be identified, the standardized target format, and the associated fusion parameters. S53. Based on the rule-parsing field mapping table, the input data is subjected to spatial benchmark unification, time format alignment, and standardized format conversion according to the standardized target format involved therein to obtain a standardized dataset; S54. Based on the standardized dataset, according to the association and fusion parameters involved in the rule-parsing field mapping table, multi-source data elements with spatial proximity relationships are associated and bound, and time-series data with different sampling frequencies are aligned to a unified time axis to obtain a spatiotemporally aligned fused dataset.

7. The method according to claim 1, characterized in that, The method further includes: S6. Based on the spatiotemporally aligned fused dataset, verification is performed from three dimensions: spatial consistency, temporal consistency, and logical consistency. The verified data is then encapsulated, and the encapsulated dataset is output as the governance result.

8. A multi-source heterogeneous spatiotemporal data governance system, characterized in that, The system includes a spatiotemporal feature analysis module, a dedicated knowledge base adaptation module, a comprehensive retrieval score calculation module, a target governance rule retrieval module, and a multi-source heterogeneous data association and fusion module, wherein: The spatiotemporal feature parsing module is used to take the acquired multi-source heterogeneous spatiotemporal data as input data and perform feature parsing on the multi-source heterogeneous spatiotemporal data to obtain spatiotemporal feature information. The dedicated knowledge base adaptation module is used to determine a dedicated knowledge base that adapts to the multi-source heterogeneous spatiotemporal data. The governance rules in the dedicated knowledge base cover at least one of spatiotemporal coordinate standards, temporal association rules and multi-source data adaptation strategies, and are bound to corresponding spatiotemporal metadata tags. The comprehensive retrieval score calculation module is used to input the spatiotemporal feature information of the input data and the spatiotemporal metadata tags bound to various governance rules in the exclusive knowledge base into the spatiotemporal attention enhancement RAG algorithm model, and perform fusion calculation based on the dynamic weights adaptively adjusted according to the differences in spatiotemporal characteristics of the data to obtain the comprehensive retrieval score. The target governance rule recall module is used to recall the Top-K target governance rules from the dedicated knowledge base based on the obtained comprehensive search scores, through rule descending sorting and spatiotemporal consistency verification. The multi-source heterogeneous data association and fusion module is used to unify the spatiotemporal benchmark of the input data and perform spatiotemporal association and fusion of multi-source heterogeneous data based on the Top-K target governance rules to obtain the fused dataset after governance. The comprehensive retrieval score calculation module performs a fusion calculation based on dynamically adjusted weights that adaptively adjust for differences in the spatiotemporal characteristics of the data, resulting in a comprehensive retrieval score, including: For each governance rule and the input data, the spatial attention score is calculated using the following formula: ； in, Indicates spatial attention score, Represents the spatial feature vector of the input data. Represents the learnable input data spatial projection matrix, used to... A linear mapping to a high-dimensional representation space. A spatial metadata vector representing governance rules. Represents a learnable regular space projection matrix, used to... A linear mapping to the same representation space as the input data. Represents the spatial feature dimension. This represents a spatial validity mask, which takes the value of 0 or 1. It is used to determine whether the spatial location of the input data is within the spatial coverage of the governance rules, ensuring that the rule score for spatial mismatch is zero. For each input data and governance rule, the time attention score is calculated using the following formula: ； in, Indicates time attention score, This represents the time feature vector of the input data, which is obtained by encoding the implicit time information in the input data. Represents the learnable input data time projection matrix, used to... Linear mapping to a high-dimensional representation space, This represents the time-based metadata vector of the governance rules, which is encoded based on the pre-marked time validity period of each rule. This represents the learnable rule-time projection matrix, used to govern the mapping of the original temporal metadata of the rules to the same high-dimensional space as the input data. Representing the time feature dimension, This represents a time validity mask, which takes the value 0 or 1. It is used to determine whether the timestamp of the input data falls within the time validity period of the governance rule, ensuring that the rule with a time mismatch receives a score of zero. The spatial attention score and the temporal attention score are weighted and fused according to a dynamic weighting calculation strategy to obtain the comprehensive retrieval score: ； ； in, This represents the overall search score, and k represents the adjustment parameter used to control the sensitivity of weight adjustments. This represents a dynamic weight value, which is dynamically adjusted based on the difference between spatial attention score and temporal attention score.

9. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a multi-source heterogeneous spatiotemporal data governance method as described in any one of claims 1-7.

Images